AAAI 2026 Papers

n the agricultural domain, the deployment of large language models (LLMs) is hindered by the lack of training data and evaluation benchmarks. To mitigate this issue, we propose AgriEval, the first comprehensive Chinese agricultural benchmark with three main characteristics: (1) Comprehensive Capability Evaluation. AgriEval covers six major agriculture categories and 29 subcategories within agriculture, addressing four core cognitive scenarios—memorization, understanding, inference, and generation. (2) High-Quality Data. The dataset is curated from university-level examinations and assignments, providing a natural and robust benchmark for assessing the capacity of LLMs to apply knowledge and make expert-like decisions. (3) Diverse Formats and Extensive Scale. AgriEval comprises 14,697 multiple-choice questions and 2,167 open-ended question-and-answer questions, establishing it as the most extensive agricultural benchmark available to date. We also present comprehensive experimental results over 51 open-source and commercial LLMs. The experimental results reveal that most existing LLMs struggle to achieve 60 percent accuracy, underscoring the developmental potential in agricultural LLMs. Additionally, we conduct extensive experiments to investigate factors influencing model performance and propose strategies for enhancement.

Hate speech detection on Chinese social networks presents distinct challenges, particularly due to the widespread use of cloaking techniques designed to evade conventional text-based detection systems. Although large language models (LLMs) have recently improved hate speech detection capabilities, the majority of existing work has concentrated on English datasets, with limited attention given to multimodal strategies in the Chinese context. In this study, we propose MMBERT, a novel BERT-based multimodal framework that integrates textual, speech, and visual modalities through a Mixture-of-Experts (MoE) architecture. To address the instability associated with directly integrating MoE into BERT-based models, we develop a progressive three-stage training paradigm. MMBERT incorporates modality-specific experts, a shared self-attention mechanism, and a router-based expert allocation strategy to enhance robustness against adversarial perturbations. Empirical results in several Chinese hate speech datasets show that MMBERT significantly surpasses fine-tuned BERT-based encoder models, fine-tuned LLMs, and LLMs utilizing in-context learning approaches.

Large Language Models (LLMs) have demonstrated strong capabilities across diverse tasks under the example-driven learning paradigm. However, in high-stakes domains such as emergency response and industrial safety, historical incidents are scarce, confidential, or both, while concise rule books are abundant. We formalize this underexplored setting as rule knowledge-driven reasoning and ask: Can LLMs reason reliably when rules are plentiful but examples are nearly absent? To study this question, we introduce RULER, an automatic benchmark that generates 32K rigorously verified questions from 1K expert-curated emergency response rules to probe three core abilities: rule memorization, single-rule application, and multi-rule complex reasoning. RULER is further equipped with a hallucination-aware evaluation suite and novel relational metrics. A comprehensive empirical study of five representative LLMs and five enhancement strategies shows that, even when models achieve reliable performance on rule memorization and single-rule application, multi-rule complex reasoning plateaus at 5.4 on a 10-point scale. To address this limitation, we propose RAMPS, a Rule knowledge-Aware Monte Carlo Tree Search Process-reward Supervision framework. RAMPS injects rule knowledge priors into MCTS, distills 12K step-level traces without human annotation, and trains an advantage-based reward model that scores candidate reasoning paths during beam search inference. Experimental results show that RAMPS significantly improves multi-rule complex reasoning performance to 7.7.

Spatial transcriptomics provides unprecedented opportunities to analyze gene patterns while preserving spatial tissue architecture. However, traditional deep learning methods for spatial transcriptomics analysis face significant challenges in multi-modal data integration, spatial dependency modeling, and biological knowledge incorporation, while existing large language models lack explicit spatial modeling capabilities for transcriptomic data. So we first present a Spatial Transcriptomics Embedding with Large Language Models (ST-LLM), a novel simple and effective approach that transforms intricate spatial graph structures into structured textual representations suitable for large language models (LLMs). ST-LLM dynamically constructs graph adjacency construction using reinforcement learning paradigms to adaptively optimize spatial relationships, converts the resulting graphs into hierarchical textual descriptions with spatial context, and leverages pre-trained semantic understanding to generate high-dimensional spatial-aware representations. Comprehensive experiments on 14 datasets demonstrate that ST-LLM achieves comparable or better performance than traditional model. ST-LLM shows that LLMs embeddings provide a new simple and effective path to encoding spatial transcriptomics biological knowledge.

Large Vision-Language Models (LVLMs) often suffer from object hallucination, making erroneous judgments about the presence of objects in images. We propose this primarily stems from spurious correlations arising when models strongly associate highly co-occurring objects during training, leading to hallucinated objects influenced by visual context. Current benchmarks mainly focus on hallucination detection but lack a formal characterization and quantitative evaluation of spurious correlations in LVLMs. To address this, we introduce causal analysis into the object recognition scenario of LVLMs, establishing a Structural Causal Model (SCM). Utilizing the language of causality, we formally define spurious correlations arising from co-occurrence bias. To quantify the influence induced by these spurious correlations, we develop Causal-HalBench, a benchmark specifically constructed with counterfactual samples and integrated with comprehensive causal metrics designed to assess model robustness against spurious correlations. Concurrently, we propose an extensible pipeline for the construction of these counterfactual samples, leveraging the capabilities of proprietary LVLMs and Text-to-Image (T2I) models for their generation. Our evaluations on mainstream LVLMs using Causal-HalBench demonstrate these models exhibit susceptibility to spurious correlations, albeit to varying extents.

Although Vision Language Models (VLMs) exhibit strong perceptual abilities and impressive visual reasoning, they struggle with attention to detail and precise action planning in complex, dynamic environments, leading to subpar performance. Real-world tasks typically require complex interactions, advanced spatial reasoning, long-term planning, and continuous strategy refinement, usually necessitating understanding the physics rules of the target scenario. However, evaluating these capabilities in real-world scenarios is often prohibitively expensive. To bridge this gap, we introduce DeepPHY, a novel benchmark framework designed to systematically evaluate VLMs' understanding and reasoning about fundamental physical principles through a series of challenging simulated environments. DeepPHY integrates multiple physical reasoning environments of varying difficulty levels and incorporates fine-grained evaluation metrics. Our evaluation finds that even state-of-the-art VLMs struggle to translate descriptive physical knowledge into precise, predictive control.

Distilling reasoning paths from teacher to student models via supervised fine-tuning (SFT) provides a shortcut for improving the reasoning ability of the smaller Large Language Models (LLMs). However, the reasoning paths generated by teacher models often reflect only surface-level traces of their underlying authentic reasoning. Insights from cognitive neuroscience suggest that authentic reasoning involves a complex interweaving between meta-reasoning that selects the appropriate sub-problem from multiple candidates, and solving, which addresses the sub-problem. It means that authentic reasoning has implicit multi-branch structure. Supervised fine-tuning collapses this rich structure into a flat sequence of token prediction in teacher's reasoning path, which cannot distill this structure to student. To address this limitation, we propose RLKD, a reinforcement learning (RL)-based distillation framework guided by a novel Generative Structure Reward Model (GSRM). Our GSRM converts the reasoning path into multiple meta-reasoning-solving steps and gives the reward to measure the alignment between the reasoning structures of student and teacher. Our RLKD combines this reward with RL, enables the student LLM to internalize the teacher’s implicit multi-branch structure in authentic reasoning, rather than merely mimicking fixed teacher's output paths. Experiments show that RLKD, even when trained on only 0.1% of the data under an RL-only regime, surpasses the performance of standard SFT-RL pipelines and further unleashes the potential reasoning ability of the student LLM than SFT-based distillation.

Efficient retrieval of external knowledge bases and web pages is crucial for enhancing the reasoning abilities of LLMs. Previous works on training LLMs to leverage external retrievers for solving complex problems have predominantly employed end-to-end reinforcement learning. However, these approaches neglect supervision over the reasoning process, making it difficult to guarantee logical coherence and rigor. To address these limitations, we propose Thinker, a hierarchical thinking model for deep search through multi-turn interaction, making the reasoning process supervisable and verifiable. It decomposes complex problems into independently solvable sub-problems, each dually represented in both natural language and an equivalent logical function to support knowledge base and web searches. Concurrently, dependencies between sub-problems are passed as parameters via these logical functions, enhancing the logical coherence of the problem-solving process. To avoid unnecessary external searches, we perform knowledge boundary determination to check if a sub-problem is within the LLM's intrinsic knowledge, allowing it to answer directly. Experimental results indicate that with as few as several hundred training samples, the performance of Thinker is competitive with established baselines. Furthermore, when scaled to the full training set, Thinker significantly outperforms these methods across various datasets and model sizes.

With the rapid advancement of large language models (LLMs), aligning them with human values for safety and ethics has become a critical challenge. This problem is especially challenging when multiple, potentially conflicting human values must be considered and balanced. Although several variants of existing alignment methods (such as Reinforcement Learning from Human Feedback (RLHF) and Direct Preference Optimization (DPO)) have been proposed to address multi-value alignment, they suffer from notable limitations: 1) they are often unstable and inefficient in multi-value optimization; and 2) they fail to effectively handle value conflicts. As a result, these approaches typically struggle to achieve optimal trade-offs when aligning multiple values. To address this challenge, we propose a novel framework called Multi-Value Alignment (MVA). It mitigates alignment degradation caused by parameter interference among diverse human values by minimizing their mutual information. Furthermore, we propose a value extrapolation strategy to efficiently explore the Pareto frontier, thereby constructing a set of LLMs with diverse value preferences. Extensive experiments demonstrate that MVA consistently outperforms existing baselines in aligning LLMs with multiple human values.

Knowledge editing aims to efficiently modify the internal knowledge of large language models (LLMs) without compromising their other capabilities. The prevailing editing paradigm, which appends an update matrix to the original parameter matrix, has been shown by some studies to damage key numerical stability indicators (such as condition number and norm), thereby reducing editing performance and general abilities, especially in sequential editing scenario. Although subsequent methods have made some improvements, they remain within the additive framework and have not fundamentally addressed this limitation. To solve this problem, we analyze it from both statistical and mathematical perspectives and conclude that multiplying the original matrix by an orthogonal matrix does not change the numerical stability of the matrix. Inspired by this, different from the previous additive editing paradigm, a multiplicative editing paradigm termed Multiplicative Orthogonal Sequential Editing (MOSE) is proposed. Specifically, we first derive the matrix update in the multiplicative form, the new knowledge is then incorporated into an orthogonal matrix, which is multiplied by the original parameter matrix. In this way, the numerical stability of the edited matrix is unchanged, thereby maintaining editing performance and general abilities. We compared MOSE with several current knowledge editing methods, systematically evaluating their impact on both editing performance and the general abilities across three different LLMs. Experimental results show that MOSE effectively limits deviations in the edited parameter matrix and maintains its numerical stability. Compared to current methods, MOSE achieves a 12.08% improvement in sequential editing performance, while retaining 95.73% of general abilities across downstream tasks.

Information retrieval (IR) systems play a critical role in navigating information overload across various applications. Existing IR benchmarks primarily focus on simple queries that are semantically analogous to single- and multi-hop relations, overlooking complex logical queries involving first-order logic operations such as conjunction (∧), disjunction (∨), and negation (¬). Thus, these benchmarks can not be used to sufficiently evaluate the performance of IR models on complex queries in real-world scenarios. To address this problem, we propose a novel method leveraging large language models (LLMs) to construct a new IR dataset ComLQ for Complex Logical Queries, which comprises 2,909 queries and 11,251 candidate passages. A key challenge in constructing the dataset lies in capturing the underlying logical structures within unstructured text. Therefore, by designing the subgraph-guided prompt with the subgraph indicator, an LLM (such as GPT-4o) is guided to generate queries with specific logical structures based on selected passages. All query-passage pairs in ComLQ are ensured structure conformity and evidence distribution through expert annotation. To better evaluate whether retrievers can handle queries with negation, we further propose a new evaluation metric, Log-Scaled Negation Consistency (LSNC@K). As a supplement to standard relevance-based metrics (such as nDCG and mAP), LSNC@K measures whether top-K retrieved passages violate negation conditions in queries. Our experimental results under zero-shot settings demonstrate existing retrieval models' limited performance on complex logical queries, especially on queries with negation, exposing their inferior capabilities of modeling exclusion. In summary, our ComLQ offers a comprehensive and fine-grained exploration, paving the way for future research on complex logical queries in IR.

Recent advances in multimodal large language models (MLLMs) have significantly improved medical AI, enabling it to unify the understanding of visual and textual information. However, as medical knowledge continues to evolve, it is critical to allow these models to efficiently update outdated or incorrect information without retraining from scratch. Although textual knowledge editing has been widely studied, there is still a lack of systematic benchmarks for multimodal medical knowledge editing involving image and text modalities. To fill this gap, we present MedMKEB, the first comprehensive benchmark designed to evaluate the reliability, generality, locality, portability, and robustness of knowledge editing in medical multimodal large language models. MedMKEB is built on a high-quality medical visual question-answering dataset and enriched with carefully constructed editing tasks, including counterfactual correction, semantic generalization, knowledge transfer, and adversarial robustness. We incorporate human expert validation to ensure the accuracy and reliability of the benchmark. Extensive experiments on state-of-the-art general and medical MLLMs demonstrate the limitations of existing knowledge editing methods in the medical domain, highlighting the need to develop specialized editing strategies.

Large Language Models (LLMs) are prone to generating incorrect or outdated information, thereby necessitating efficient and precise mechanisms for knowledge updates. Existing knowledge editing approaches, however, often encounter conflicts between two competing objectives: maintaining existing knowledge (preservation) and incorporating new information (editing). During gradient-based optimization, these conflicting objectives can lead to imbalanced update directions, where one gradient dominates, ultimately resulting in suboptimal learning dynamics. To address this challenge, we propose a balanced knowledge editing framework inspired by Nash bargaining theory. Our method guides the optimization process toward a Pareto stationary point, ensuring an equilibrium solution wherein any deviation from the final state would degrade the overall performance with respect to both objectives. This guarantees optimality in preserving prior knowledge while integrating new information. We empirically validate the effectiveness of our approach across a range of evaluation metrics on standard benchmark datasets. Extensive experiments show that our method consistently outperforms state-of-the-art techniques, achieving a superior balance between knowledge preservation and update accuracy.

Low-Rank Adaptation (LoRA) enables efficient fine-tuning of large language models but suffers from catastrophic forgetting when learned updates interfere with the dominant singular directions that encode essential pre-trained knowledge. We propose Orthogonal Projection LoRA (OPLoRA), a theoretically grounded approach that prevents this interference through double-sided orthogonal projections. By decomposing frozen weights via SVD, OPLoRA constrains LoRA updates to lie entirely within the orthogonal complement of the top-k singular subspace using projections PL = I − Uk Ukᵀ and PR = I − Vk Vkᵀ. We prove that this construction exactly preserves the top-k singular triples, providing mathematical guarantees for knowledge retention. To quantify subspace interference, we introduce ρk, a metric measuring update alignment with dominant directions. Extensive experiments across commonsense reasoning, mathematics, and code generation demonstrate that OPLoRA significantly reduces forgetting while maintaining competitive task-specific performance on LLaMA-2 7B and Qwen2.5 7B, establishing orthogonal projection as an effective mechanism for knowledge preservation in parameter-efficient fine-tuning.

Black-box tuning is an emerging paradigm for adapting large language models (LLMs) to better achieve desired behaviors, particularly when direct access to model parameters is unavailable. Current strategies, however, often present a dilemma of suboptimal extremes: either separately train a small proxy model and then use it to shift the predictions of the foundation model, offering notable efficiency but often yielding limited improvement; or making API calls in each tuning iteration to the foundation model, which entails prohibitive computational costs. In this paper, we argue that a more reasonable way for black-box tuning is to train the proxy model with limited API calls. The underlying intuition is based on two key observations: first, the training samples may exhibit correlations and redundancies, suggesting that the foundation model’s predictions can be estimated from previous calls; second, foundation models frequently demonstrate low accuracy on downstream tasks. Therefore, we propose a novel advanced black-box tuning method for LLMs with limited API calls. Our core strategy involves training a Gaussian Process (GP) surrogate model with "LogitMap Pairs" derived from querying the foundation model on a minimal but highly informative training subset. This surrogate can approximate the outputs of the foundation model to guide the training of the proxy model, thereby effectively reducing the need for direct queries to the foundation model. Extensive experiments verify that our approach elevates pre-trained language model accuracy from 55.92% to 86.85%, reducing the frequency of API queries to merely 1.38%. This significantly outperforms offline approaches that operate entirely without API access. Notably, our method also achieves comparable or superior accuracy to query-intensive approaches, while significantly reducing API costs. This offers a robust and high-efficiency paradigm for language model adaptation.

Knowledge Distillation (KD) is a key technique for compressing Large-scale Language Models (LLMs), but prevailing logit-based methods employ static strategies misaligned with the student’s dynamic learning process. By treating all tokens indiscriminately with a fixed temperature, these methods result in suboptimal knowledge transfer. To address this, we propose LLM-oriented token-Adaptive Knowledge Distillation (AdaKD), a framework that adapts the distillation process to each token’s real-time learning state. AdaKD consists of two synergistic modules driven by a unified token difficulty metric. First, the Loss-driven Adaptive Token Focusing (LATF) module dynamically concentrates distillation on valuable tokens by monitoring the student’s learning stability. Second, Inverse Difficulty Temperature Scaling (IDTS) introduces a counterintuitive token-level temperature: low for difficult tokens to target error correction, and high for easy tokens to learn the teacher’s smooth output distribution for better generalization. As a plug-and-play framework, AdaKD consistently improves performance across diverse distillation methods, model architectures, and benchmarks.

Dense retrieval has become a foundational paradigm in modern search systems, especially on short-video platforms. However, most industrial systems adopt a self-reinforcing training pipeline that relies on historically exposed user interactions for supervision. This paradigm inevitably leads to a filter bubble effect, where potentially relevant but previously unseen content is excluded from the training signal, biasing the model toward narrow and conservative retrieval. In this paper, we present CroPS (Cross-Perspective Positive Samples), a novel retrieval data engine designed to alleviate this problem by introducing diverse and semantically meaningful positive examples from multiple perspectives. CroPS enhances training with positive signals derived from user query reformulation behavior (query-level), engagement data in recommendation streams (system-level), and world knowledge synthesized by large language models (knowledge-level). To effectively utilize these heterogeneous signals, we introduce a Hierarchical Label Assignment (HLA) strategy and a corresponding H-InfoNCE loss that together enable fine-grained, relevance-aware optimization. Extensive experiments conducted on Kuaishou Search, a large-scale commercial short-video search platform, demonstrate that CroPS significantly outperforms strong baselines both offline and in live A/B tests, achieving superior retrieval performance and reducing query reformulation rates. CroPS is now fully deployed in Kuaishou Search, serving hundreds of millions of users daily.

Optimization modeling plays a critical role in supporting optimal decision-making across various domains. Previous works have demonstrated that large language models (LLMs) tailored for optimization modeling have significantly automated and simplified this process. However, these models typically employ a straightforward input-output paradigm and struggle with challenging instances. In contrast, recent advances in general-purpose reasoning LLMs (RLLMs), such as DeepSeek-R1, have shown impressive capabilities in complex domains like mathematics and coding. In this paper, we introduce DeepOR, the first RLLM specifically designed for optimization modeling. Instead of directly outputting solutions, DeepOR explicitly performs multiple intermediate reasoning steps. To adapt a base LLM into an RLLM, we begin by synthesizing long chain-of-thought (CoT) data guided by a flowchart, which is automatically generated using a self-exploration algorithm. Once the training data are prepared, we employ supervised fine-tuning on the base LLM to endow it with reasoning capabilities tailored for optimization modeling. To fully leverage the model's reasoning potential, we further apply reinforcement learning with reward-shaping derived from solver feedback. Experimental results on benchmarks confirm that DeepOR consistently and significantly outperforms existing state-of-the-art approaches.

Large Language Models (LLMs) have achieved remarkable success across a wide range of natural language tasks, but often exhibit overconfidence and generate plausible yet incorrect answers. This overconfidence, especially in models undergone Reinforcement Learning from Human Feedback (RLHF), poses significant challenges for reliable uncertainty estimation and safe deployment. In this paper, we propose EAGLE (Expectation of AGgregated internaL bEief), a novel self-evaluation-based calibration method that leverages the internal hidden states of LLMs to derive more accurate confidence scores. Instead of relying on the model's final output, our approach extracts internal beliefs from multiple intermediate layers during self-evaluation. By aggregating these layer-wise beliefs and calculating the expectation over the resulting confidence score distribution, EAGLE produces a refined confidence score that more faithfully reflects the model's internal certainty. Extensive experiments on diverse datasets and LLMs demonstrate that EAGLE significantly improves calibration performance over existing baselines. We also provide an in-depth analysis of EAGLE, including a layer-wise examination of uncertainty patterns, a study of the impact of self-evaluation prompts, and an analysis of the effect of self-evaluation score range.